Product · Pillar II · Protect
Guardrails that enforce your policy, not a generic one.
PromptGuard screens text. MediaGuard screens images. Both enforce bespoke policies on any data, AI-generated or not, and return a verdict with a threat score and a rationale, in production, on every check.
Non-Compliant· threat 0.94check_media →Compliant· threat 0.06PromptGuard for text. MediaGuard for images.
PromptGuard
Text, screened against your rules
User prompts, model outputs, agent messages, documents. Anything textual is screened for the policies you define: prohibited topics, regulated claims, brand rules, disclosure risks, and anything else your reviewers require, in real time on production traffic.
POST /v1/guardrails/promptMediaGuard
Images, screened with the same rigor
Generated images and uploads, checked against bespoke visual policies: content categories, composition rules, brand and platform requirements. Same policy model, same API, in production.
POST /v1/guardrails/mediaHundreds of rules. One flat price.
Generic moderation categories are where policy goes to die. Guardrails run your policy, written with your team, versioned, and enforced verbatim. Customers configure up to 225 custom policy rules today, and pricing stays flat no matter how many you run.
Bespoke by default
Policies express your obligations (network rules, regulatory language, platform terms), not a vendor's fixed taxonomy.
Any data
AI-generated or human-written, inbound or outbound, text or image: if it flows through your product, it can be checked.
Explained verdicts
Every check returns a compliance status, a threat level, and a rationale, so your logs answer "why" before anyone asks.
Measured in production, not on a slide.
Every figure below is measured from live production traffic.
Built to be hard to break.
A jury, not a judge.
Each check is adjudicated by multiple models. No single model failure, provider outage, or bad response decides a verdict on its own.
Degrades in layers.
Redundant components back each other up, so a struggling dependency narrows the system instead of stopping it.
Continuity, on the record.
Served production traffic every hour for the last 90 days. A measured record, not an SLA.
Verdicts that explain themselves.
Each check is a full policy adjudication: verdict, threat score, and written rationale, not a single-classifier score.
Evaluating Bedrock Guardrails or Hive AI?
Good. Run the comparison. AetherLab has been chosen over Amazon Bedrock Guardrails and Hive AI in head-to-head evaluations in high-stakes workflows. The pattern behind those decisions:
Policy fidelity
High-stakes workflows fail on the policies a fixed category list can't express. Bedrock Guardrails caps denied topics at 30 per guardrail; AetherLab customers run bespoke rule sets more than seven times that size.
Image parity
Text and image checks share one policy model and one API. Image is not a separate, weaker product.
Cost shape
Flat pricing regardless of policy count: adding your 200th rule costs the same as your 5th, which changes how thoroughly teams are willing to write policy.
We're glad to support a structured evaluation against your current stack. Request one here.
Live in an afternoon.
Guardrails are one pip install aetherlab away. The same API screens text with check_prompt and images with check_media, against the policies you define.
from aetherlab import AetherLabClient
client = AetherLabClient() # reads AETHERLAB_API_KEY
result = client.check_prompt(
"user or model text to screen",
blacklisted_keywords=["guaranteed returns"],
)
print(result.compliance_status) # "Non-Compliant"
print(result.avg_threat_level) # 0.94Put your policy in production.
Bring your rules, or the findings from an AdversarialScan, and we'll configure guardrails against them. Flat pricing, text and image, one API.
Ask about the Evidence Pack
Leave your email and we'll walk you through what an Evidence Pack contains for your use case: severity-scored findings, business-impact mapping, and the approval record.